Search

• The paper proposes a methodology for a hybrid model that can capture complex non-linear relationships and help understand the interactions between input and output features. • The model overcomes the limitation of traditional neural network models in predicting across different spaces. • The proposed model is an intelligent water quality prediction model that can provide decision support for predicting COD Mn concentration and managing watershed risks in the Changdang Lake basin, with good application prospects and practical value. The organic pollution of lake water can cause a tremendous threat to the water ecosystem and human health. The COD Mn is one of the crucial indicators of lake water quality and is commonly utilized to gauge the extent of organic pollution in lake water. Therefore, this paper selected COD Mn as the research object and used the water quality monitoring data of Changdang Lake in China and its upstream and downstream to predict the COD Mn concentration in the lake. In order to study the spatial relationship between the lake and upstream and downstream water quality, reflect the joint action of multiple water quality factors in prediction and the interaction between different feature factors. This study combined the XGBoost feature filtering algorithm, maximum mutual information coefficient (MIC), and improved recurrent neural network (GRU) and proposes a hybrid model called XGB-MIC-GRU. The model first used XGBoost to screen and extract the relative importance of water quality characteristics and used the Shapley addition extension (SHAP) method to explain XGBoost feature extraction. Then, the correlation between the lake and the upstream and downstream water quality is calculated through MIC analysis. Finally, the selected water quality factor characteristics and spatial characteristics are input into the GRU model for prediction. The experimental results showed that water temperature, total phosphorus, and total nitrogen are the most important to COD Mn , and the upstream US1 and downstream DS1 and DS2 stations are the most closely related to the concentration of COD Mn in the lake. By comparing the prediction effect of the model in different time steps, the best 16-time steps related data were selected to predict the value of the next time. MAE, RMSE, and R2 of the model are 0.10, 0.13, and 0.96, respectively. The model has better prediction accuracy and correlation error than the traditional SVR and GPR. The proposed mixed model can accurately predict the concentration of COD Mn in the lake. It can assist decision-makers in timely implementation of effective measures to safeguard the lake ecosystem. [ABSTRACT FROM AUTHOR]

• Sensor technology developments drove advances in biophysical estimation techniques. • CubeSat constellations will increase VHR data accessibility to various users. • Sentinel-2 red-edge bands present prospects for operational crop stress detection. • RF is prominently used in the literature despite new developments in algorithms. • Retrieval models developed in Mediterranean areas do not apply to semi-arid areas. This paper provides an overview of the recent developments in remote sensing technology and machine learning algorithms for estimating important biophysical and biochemical parameters for precision farming. The objectives are (i) to provide an overview of recent advances in remotely sensed retrieval of biophysical and biochemical parameters brought by the developments in sensor technologies and robust machine learning algorithms and (ii) to identify the sources of uncertainty in retrieving biophysical and biochemical parameters and implications for precision agriculture. The review revealed that developments in crop biophysical and biochemical parameters retrieval techniques were mainly driven by announcements and the availability of new sensors. Two ground-breaking events can be identified, i.e., the availability of Sentinel-2 and the SuperDove constellation. The two provide high temporal-high spatial resolution data relevant for site-specific management and super-spectral configuration, enabling retrieval of crop growth and health parameters. The free availability of Sentinel-2 triggered the testing of its spectral configurations and upscaling of retrieval approaches using simulated data from field spectrometers and airborne hyperspectral sensors. SuperDoves will likely reduce the cost of very high-resolution data while providing unprecedented capabilities for detailed, accurate and frequent characterisation of field variability. Studies showed that the red-edge bands and hybrid models coupling Radiative Transfer Model (RTM) and machine learning regression algorithms (MLRA) are promising for operational and accurate monitoring of stress-related crop parameters to aid time-sensitive agronomic decisions. However, such models were tested in Mediterranean climates and performed poorly in African semi-arid areas and China's temperate continental semi-humid monsoon climates. Therefore, locally-calibrated RTM models incorporating crop-type maps and other spatio-temporal constraints may reduce uncertainties when adapted to data-scarce regions. Generally, permanent experimental sites and a lack of systematic calibration data on various crops are some limiting factors to using remote sensing technologies for PA in Sub-Saharan Africa. Other complexities arise from farm configurations, such as small field sizes and mixed cropping practices. Therefore, future studies should develop generic, scalable and transferable models, especially within under-studied areas. [ABSTRACT FROM AUTHOR]

• Machine learning was applied to quantify the effects of stand, climate and soil on biomass of mixed forests with good accuracy. • Quantifying the importance of various environmental attribute factor to forest biomass depends on the combination of environmental variables and modeling methods. • Environmental variables including degree-days above 18 °C, bulk density and soil pH were most important predictors. Forest biomass is influenced by multiple environmental factors. Multi-source data on site-specific soil, climate, and stand factors are essential for stand biomass prediction. While previous studies concerning planted forest biomass at stand level highlighted the importance of environmental factors, few have been conducted on the effects of climate and soil on stand biomass estimation, especially for natural mixed forest with complicated stand structure. Machine learning (ML) algorithms provide new tools for examining the effects, but the applications are still limited. Therefore, the objectives of this paper were to develop ML models for stand biomass estimation with stand and multiple environmental predictors, and quantify the relative importance of environmental variables. Tree above- and below-ground biomass at stand level was set as output variable with the random forest (RF) and boosted regression tree (BRT) regression methods. The data were from 286 sample plots from the National Forest Inventory (NFI) in Jilin Province, northeast China. The results showed that models with the inclusion of stand, soil, and climate variables explained > 89% of forest biomass variations. BA (stand basal area) and H (stand average height) were the most important stand variables, DD18 (degree-days above 18 °C) was the most important climate variable, and bdod (bulk density) and pH were the most important soil variables. Relative importance of climate and soil factors was 0.5% to 5.1% and 1.2% to 4.1%, respectively, which were weak but important predictors of stand biomass. In the RF model, climate variables were more important than soil variables, but the opposite in the BRT models. We concluded that above forest stand, climate and soil factors were significant input variables for regional forest biomass mapping using ML. [ABSTRACT FROM AUTHOR]

• An improved adaptive updating ant colony optimization algorithm (AU-ACO) is proposed. • Feature bands extracted by AU-ACO is taken as input to reduce the complexity of model. • A new hybrid prediction model (ELM-P) combining ELM with PLSR is proposed. • This model can estimate rice chlorophyll content with extracted feature band. • The propose method has better superiority compared with other modeling methods. Using spectral information to detect chlorophyll content in rice canopy leaves quickly, non-destructively and accurately has a great practical significance for rice growth evaluation, precise fertilization and scientific management. In this paper, Japonica rice in northeast China is taken as the research object, and rice canopy hyperspectral data of key growth stages are obtained through plot experiments. Firstly, the standard normal variate (SNV) and SG smoothing method is used to preprocess the hyperspectral data, and based on the processed spectral data and ant colony optimization algorithm (ACO), an improved adaptive updating ant colony optimization algorithm (AU-ACO) is proposed to select feature bands of chlorophyll content by introducing an adaptive adjustment strategy of volatilization coefficient and optimal pheromone updating strategy in different stages, and compared with the standard ACO algorithm and full-band modeling method. Then, taking the extracted feature band and full-band as input, considering the advantages of linear model and nonlinear model, a hybrid prediction model (ELM-P) combining extreme learning machine (ELM) with partial least squares regression (PLSR) is proposed. In this model, PLSR is used to obtain the preliminary prediction of the chlorophyll content in rice, and the linear trend is obtained, and then the ELM with strong nonlinear approximation ability is used to predict the deviation of PLSR model, the final prediction value is obtained by superposition of the two outputs. In order to verify the superiority of the proposed model, PLSR and ELM prediction models are also established by taking the full-band and the feature bands by different extraction methods as input. The simulation experiment results show that under the same prediction model conditions, the feature bands extracted by proposed AU-ACO algorithm as input can reduce the complexity of the model and improve the prediction performance of the model. The determination coefficient (R 2- P) of testing set and the determination coefficient (R 2- C) of training set for each model are greater than 0.6785, among them, the ELM-P model with the feature band extracted by AU-ACO algorithm as input has the highest prediction accuracy, R 2- C and R 2- P are 0.7969 and 0.7918 respectively, RMSE- C and RMSE- P are 1.2969 and 1.1293 mg/L respectively, which can provide valuable reference for detecting and evaluating chlorophyll content in Japonica rice. [ABSTRACT FROM AUTHOR]